论文部分内容阅读
In this paper, we propose a coarse-to-fine convolutional network framework designed with problem specific knowledge for fast automatic portrait segmentation. We built up a dataset of 7 100 portrait images which are frames from personal live show videos. T